taxonomy of scala

A Taxonomy of ScalaStrangeLoop 2012

Jamie Allen@jamie_allen

http://github.com/jamie-allen/taxonomy-of-scala

Agenda

• Goal• Object-Oriented Features• Pattern Matching• Functional Programming• Actors• Futures• Implicits• Type Theory• Macros• Category Theory

Goal

Provide you with a reference point for many of the terms you hear in the Scala community

How Programming in Scala Makes Me Feel

How I Write Programs

• Pre-Scala:– Make it work– Make it work well– Make it work fast

• With Scala:– Make it work and work well– Make it work fast

Object-Oriented Features

Case Classes

case class Person(firstName: String = "Jamie", lastName: String = "Allen")

val jamieDoe = Person(lastName = "Doe") res0: Person = Person(Jamie,Doe)

• Data Transfer Objects (DTOs) done right• By default, class arguments are immutable & public• Should never be extended• Provide equals(), copy(), hashCode() and toString()

implementations• Don’t have to use new keyword to create instances• Named Parameters and Default arguments give us Builder pattern semantics

Lazy Definitionslazy val calculatedValue = piToOneMillionDecimalPoints()

• Excellent for deferring expensive operations until they are needed

• Reducing initial footprint• Resolving ordering issues• Implemented with a guard field and

synchronization, ensuring it is created when necessary

Importsimport scala.collection.immutable.Map

class Person(val fName: String, val lName: String) {import scala.collection.mutable.{Map => MMap}val cars: MMap[String, String] = MMap()...

}

• Can be anywhere in a class• Allow for selecting multiple classes from a package or using

wildcards• Aliasing• Order matters!

Objectsobject Bootstrapper extends App { Person.createJamieAllen }

object Person { def createJamieAllen = new Person("Jamie", "Allen") def createJamieDoe = new Person("Jamie", "Doe") val aConstantValue = "A constant value”}

class Person(val firstName: String, val lastName: String)

• Singletons within a JVM process• No private constructor histrionics• Companion Objects, used for factories and constants

The apply() methodArray(1, 2, 3)res0: Array[Int] = Array(1, 2, 3)

res0(1)res1: Int = 2

• In companion objects, it defines default behavior if no method is called on it

• In a class, it defines the same thing on an instance of the class

Tuplesdef firstPerson = (1, Person(firstName = “Barbara”))val (num: Int, person: Person) = firstPerson

• Binds you to an implementation• Great way to group values without a DTO• How to return multiple values, but wrapped in

a single instance that you can bind to specific values

Pattern Matching

Pattern Matching Examplesname match { case "Lisa" => println("Found Lisa”) case Person("Bob") => println("Found Bob”) case "Karen" | "Michelle" => println("Found Karen or Michelle”) case Seq("Dave", "John") => println("Got Dave before John”) case Seq("Dave", "John", _*) => println("Got Dave before John”) case ("Susan", "Steve") => println("Got Susan and Steve”) case x: Int if x > 5 => println("got value greater than 5: " + x) case x => println("Got something that wasn't an Int: " + x) case _ => println("Not found”)}

• A gateway drug for Scala• Extremely powerful and readable• Not compiled down to lookup/table switch unless you

use the @switch annotation,

Functional Programming

Immutability

• Extends beyond marking instances final• You must not leak mutability

Referential Transparency// Transparentval example1 = "jamie".reverseval example2 = example1.reverseprintln(example1 + example2) // eimajjamie

// Opaqueval example1 = new StringBuffer("Jamie").reverseval example2 = example1.reverseprintln(example1 append example2) // jamiejamie

• An expression is transparent if it can be replaced by its VALUE without changing the behavior of the program

• In math, all functions are referentially transparent

Scala Collectionsval myMap = Map(1 -> "one", 2 -> "two", 3 -> "three")val mySet = Set(1, 4, 2, 8)val myList = List(1, 2, 8, 3, 3, 4)val myVector = Vector(1, 2, 3...)

• You have the choice of mutable or immutable collection instances, immutable by default

• Rich implementations, extremely flexible

Rich Collection Functionalityval numbers = 1 to 20 // Range(1, 2, 3, ... 20)

numbers.head // Int = 1numbers.tail // Range(2, 3, 4, ... 20)numbers.take(5) // Range(1, 2, 3, 4, 5)numbers.drop(5) // Range(6, 7, 8, ... 20)

• There are many methods available to you in the Scala collections library

• Spend 5 minutes every day going over the ScalaDoc for one collection class

Higher Order Functionsval names = List("Barb", "May", "Jon")

names map(_.toUpperCase)res0: List[java.lang.String] = List(BARB, MAY, JON)

• Really methods in Scala• Applying closures to collections

Higher Order Functionsval names = List("Barb", "May", "Jon")

names map(_.toUpperCase)res0: List[java.lang.String] = List(BARB, MAY, JON)

names flatMap(_.toUpperCase)res1: List[Char] = List(B, A, R, B, M, A, Y, J, O, N)

names filter (_.contains("a"))res2: List[java.lang.String] = List(Barb, May)

val numbers = 1 to 20 // Range(1, 2, 3, ... 20)

numbers.groupBy(_ % 3)res3: Map[Int, IndexedSeq[Int]] = Map(1 -> Vector(1, 4, 7, 10, 13, 16, 19), 2 -> Vector(2, 5, 8, 11, 14, 17, 20), 0 -> Vector(3, 6, 9, 12, 15, 18))

For Comprehensions

• Used for composing higher-order functions• As you chain higher-order functions, you may

find it easier to reason about them this way

val myNums = 1 to 20

for (i <- myNums) yield i + 1myNums map(_ + 1)

for { i <- myNums j <- 1 to i} yield i * jmyNums flatMap(i => 1 to i map (j => i * j))

Parallel Collectionsscala> 1 to 1000000res0: scala.collection.immutable.Range.Inclusive = Range(1, 2, 3,...

scala> res0.parres1: s.c.parallel.immutable.ParRange = ParRange(1, 2, 3,...

scala> res1 map(_ + 1)res2: s.c.parallel.immutable.ParSeq[Int] = ParVector(2, 3, 4,...

scala> res2.seqres3: s.c.immutable.Range = Range(2, 3, 4,...

• You can easily parallelize the application of a function literal to your collection by calling the par() method on a collection instance

• Uses JSR166 under the covers to fork/join for you• Use the seq() method on the parallel collection to return to a non-parallel

instance

Partial Functions

• A simple match without the match keyword• The receive block in Akka actors is an excellent example• Is characterized by what "isDefinedAt" in the case

statements

class MyActor extends Actor { def receive = { case s: String => println("Got a String: " + s) case i: Int => println("Got an Int: " + i) case x => println("Got something else: " + x) }}

Curryingdef product(i: Int)(j: Int) = i * j val doubler = product(2)_doubler(3) // Int = 6doubler(4) // Int = 8

val tripler = product(3)_tripler(4) // Int = 12tripler(5) // Int = 15

• Take a function that takes n parameters as separate argument lists• “Curry” it to create a new function that only takes one parameter• Fix on a value and use it to apply a specific implementation of a

product with semantic value• Have to be defined explicitly as such in Scala• The _ is what explicitly marks this as curried

Actors

Actorsimport akka.actor._

class MyActor extends Actor { def receive = { case x => println(“Got value: “ + x) }}

• Based on concepts from Erlang/OTP• Akka is replacing the core language actors• Concurrency paradigm using networks of

independent objects that only communicate via messaging and mailboxes

Futures

Futuresimport scala.concurrent._

val costInDollars = Future { webServiceProxy.getCostInDollars.mapTo[Int]}

costInDollars map (myPurchase.setCostInDollars(_))

• Allows you to write asynchronous code, which can be more performant than blocking

• Are not typed, hence the mapTo call above

Futures in Sequenceval customerPurchases = for ( costUSD <- Future{ proxy.getCostInDollars.mapTo[Int]} totalPurchase <- Future{ proxy.addToTotal(costUSD).mapTo[Int]}} yield ((customerId -> totalPurchase))

• Scala’s for comprehensions allow you to compose higher-order functions, including Futures

• By sequencing the expressions on multiple lines, you can order dependencies

Futures in Parallelval costUSD = Future{proxy.getCostInUSD(cost).mapTo[Int]}val costCAD = Future{proxy.getCostInCAD(cost).mapTo[Int]}val combinedCosts = for { cUSD <- costUSD cCAD <- costCAD} yield (cUSD, cCAD)

val costs = for ( (costUSD, costCAD) <- Future{proxy.getCostInUSD(cost).mapTo[Int]} zip Future{proxy.getCostInCAD(cost).mapTo[Int]}} yield (costUSD, costCAD)

• Define the futures separately and then compose• Alternatively, the zip method allows you to parallelize

futures execution within a for comprehension

Implicits

Implicit Conversions

Implicit Conversionscase class Person(firstName: String, lastName: String)implicit def PersonToInt(p: Person) = p.toString.head.toInt

val me = Person("Jamie", "Allen")

val weird = 1 + me res0: Int = 81

• Looks for definitions at compile time that will satisfy type incompatibilities

• Modern IDEs will warn you with an underline when they are in use

• Limit scope as much as possible (see Josh Suereth's NE Scala 2011)

Implicit Parametersdef executeFutureWithTimeout(f: Future)(implicit t: Timeout)

implicit val t: Timeout = Timeout(20, TimeUnit.MILLISECONDS)executeFutureWithTimeout(Future {proxy.getCustomer(id)})

• Allow you to define default parameter values that are only overridden if you do so explicitly

• Handy to avoid code duplication

Implicit Classesimplicit class Person(name: String)

class Person(name: String)implicit final def Person(name: String): Person = new Person(name)

• New to Scala 2.10• Create extension methods to existing types• Desugars at compile time into a class

definition with an implicit conversion

Type Theory

Type Inference

• Declaring a variable/value• Return types of methods/functions• See Daniel Spiewak's Philly ETE 2011 talk• Good idea to show types on public interfaces• Specify types when you want to type certainty

Type Classes I

• Allow you to layer in varying implementations of behavior without changing an existing inheritance structure

case class Customer(id: Long, firstName: String, lastName: String)

trait CustomerOrderById extends Ordering[Customer] { def compare(x: Customer, y: Customer): Int = { ... }}implicit object CustomerIdSort extends CustomerOrderById

val customers = List(Customer(1, "Jamie", "Allen"), Customer(5, "John", "Doe"), Customer(2, "Jane", "Smith"))val sortedCustomers = customers.sorted(CustomerIdSort)sortedCustomers: List[Customer] = List(Customer(1,Jamie,Allen), Customer(2,Jane,Smith), Customer(5,John,Doe))

Type Classes II

• Allows you to generalize types that are acceptable parameters for methods

case class Dog(name: String)case class Ferret(name: String)case class Cat(name: String)abstract class OkayPets[T]object OkayPets { implicit object OkayDog extends OkayPets[Dog] implicit object OkayFerret extends OkayPets[Ferret]}def getPet[T](t: T)(implicit p: OkayPets[T]) = t

val myDog = getPet(Dog("Sparky")) // Worksval myCat = getPet(Cat("Sneezy")) // Fails at compile time

Higher Kinded TypesMap[A, B] // Type constructor, not a type!

val myMap = Map[Int, String]() // Now it’s a type!

• Use other types to construct a new type• Also called type constructors

Algebraic Data Typessealed abstract class DayOfTheWeekcase object Sunday extends DayOfTheWeekcase object Monday extends DayOfTheWeek ...case object Saturday extends DayOfTheWeek

val nextDay(d: DayOfTheWeek): DayOfTheWeek = d match { case Sunday => Monday case Monday => Tuesday ... case Saturday => Sunday }}

• Allow you to model the world in finite terms, such as enumerations, but also define behavior around them, with all of the power of case classes

• A finite number of possible subtypes, enforced by the "sealed" keyword (must be defined in the same source file)

Macros

Macros• New to Scala 2.10• Macros are used for generating code at

compile time, similar to LISP macros• Does not have compiler pragmas such as #ifdef

• Are implemented as "hygenic" macros at the point you call reify() – identifiers cannot be closed over in a macro definition

ScalaLogging Macrodef debug(message: String): Unit = macro LoggerMacros.debugprivate object LoggerMacros { def debug(c: LoggerContext)(message: c.Expr[String]) = c.universe.reify( if (c.prefix.splice.underlying.isDebugEnabled) c.prefix.splice.underlying.debug(message.splice) )}

import com.typesafe.scalalogging.Loggingclass MyClass extends Logging { logger.debug("This won't occur if debug is not defined")}

• Existing log libraries allow us to define logging statements and then determine whether they result in output at runtime

• ScalaLogging allows a user to use a logging facility but decide at compile time whether or not to include the logging statement based on log level.

Category Theory

Category Theory

Concepts and Arrowsval myIntToStringArrow: Int => String = _.toString

myIntToStringArrow(1100)res0: String = 1100

• Concepts are types• Arrows are functions that convert one concept

to another

Morphismval number = 1000val numericString = number.toString

• Morphisms change one value in a category to another in the same category, from one type to another where types are the category

• Simplified, it converts a type with one property to a type with another property

• Must be pure, not side-effecting

Functorval numbers = List(1, 2, 3, 4)val numericStrings = numbers.map(_.toString)

• Functors are transformations from one category to another that preserve morphisms

• Simplified, converts a type from one to another while maintaining the conversion of a type with one property to a type with another property

Monadval customerPurchases = for ( costUSD <- proxy.getCostInDollars totalPurchase <- proxy.addToTotal(costUSD)} yield ((customerId -> totalPurchase))

• Very ephemeral concept• Must meet the laws of a monad to be one• Combine functor applications because they can be bound

together, sequencing operations on the underlying types• flatMap() is the method the Scala compiler uses to

bind monads

Thank You!

Credits• Sources

– Fast Track to Scala courseware by Typesafe– Scala in Depth, by Josh Suereth– DSLs in Action, Debasish Ghosh– Wikipedia– Runar Bjarnason's NE Scala 2011 talk– Daniel Sobral's blog– Brendan McAdams' blog

• Contributors– Dave Esterkin, Chariot Solutions– Josh Suereth, Typesafe